net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer (maintainer)   <bdschuym@pandora.be>
   8  *
   9  *      Changes:
  10  *      Apr 29 2003: physdev module support (bdschuym)
  11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
  12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
  13  *                   (bdschuym)
  14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
  15  *
  16  *      This program is free software; you can redistribute it and/or
  17  *      modify it under the terms of the GNU General Public License
  18  *      as published by the Free Software Foundation; either version
  19  *      2 of the License, or (at your option) any later version.
  20  *
  21  *      Lennert dedicates this file to Kerstin Wurdinger.
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/ip.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/netfilter_bridge.h>
  33 #include <linux/netfilter_ipv4.h>
  34 #include <linux/netfilter_ipv6.h>
  35 #include <linux/netfilter_arp.h>
  36 #include <linux/in_route.h>
  37 #include <linux/inetdevice.h>
  38
  39 #include <net/ip.h>
  40 #include <net/ipv6.h>
  41 #include <net/route.h>
  42
  43 #include <asm/uaccess.h>
  44 #include "br_private.h"
  45 #ifdef CONFIG_SYSCTL
  46 #include <linux/sysctl.h>
  47 #endif
  48
  49 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  50                                  (skb->nf_bridge->data))->daddr.ipv4)
  51 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  52 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  53
  54 #ifdef CONFIG_SYSCTL
  55 static struct ctl_table_header *brnf_sysctl_header;
  56 static int brnf_call_iptables __read_mostly = 1;
  57 static int brnf_call_ip6tables __read_mostly = 1;
  58 static int brnf_call_arptables __read_mostly = 1;
  59 static int brnf_filter_vlan_tagged __read_mostly = 1;
  60 #else
  61 #define brnf_filter_vlan_tagged 1
  62 #endif
  63
  64 static inline __be16 vlan_proto(const struct sk_buff *skb)
  65 {
  66         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  67 }
  68
  69 #define IS_VLAN_IP(skb) \
  70         (skb->protocol == htons(ETH_P_8021Q) && \
  71          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  72          brnf_filter_vlan_tagged)
  73
  74 #define IS_VLAN_IPV6(skb) \
  75         (skb->protocol == htons(ETH_P_8021Q) && \
  76          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_ARP(skb) \
  80         (skb->protocol == htons(ETH_P_8021Q) && \
  81          vlan_proto(skb) == htons(ETH_P_ARP) && \
  82          brnf_filter_vlan_tagged)
  83
  84 /* We need these fake structures to make netfilter happy --
  85  * lots of places assume that skb->dst != NULL, which isn't
  86  * all that unreasonable.
  87  *
  88  * Currently, we fill in the PMTU entry because netfilter
  89  * refragmentation needs it, and the rt_flags entry because
  90  * ipt_REJECT needs it.  Future netfilter modules might
  91  * require us to fill additional fields. */
  92 static struct net_device __fake_net_device = {
  93         .hard_header_len        = ETH_HLEN
  94 };
  95
  96 static struct rtable __fake_rtable = {
  97         .u = {
  98                 .dst = {
  99                         .__refcnt               = ATOMIC_INIT(1),
 100                         .dev                    = &__fake_net_device,
 101                         .path                   = &__fake_rtable.u.dst,
 102                         .metrics                = {[RTAX_MTU - 1] = 1500},
 103                         .flags                  = DST_NOXFRM,
 104                 }
 105         },
 106         .rt_flags       = 0,
 107 };
 108
 109 static inline struct net_device *bridge_parent(const struct net_device *dev)
 110 {
 111         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 112
 113         return port ? port->br->dev : NULL;
 114 }
 115
 116 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 117 {
 118         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 119         if (likely(skb->nf_bridge))
 120                 atomic_set(&(skb->nf_bridge->use), 1);
 121
 122         return skb->nf_bridge;
 123 }
 124
 125 static inline void nf_bridge_save_header(struct sk_buff *skb)
 126 {
 127         int header_size = ETH_HLEN;
 128
 129         if (skb->protocol == htons(ETH_P_8021Q))
 130                 header_size += VLAN_HLEN;
 131
 132         skb_copy_from_linear_data_offset(skb, -header_size,
 133                                          skb->nf_bridge->data, header_size);
 134 }
 135
 136 /*
 137  * When forwarding bridge frames, we save a copy of the original
 138  * header before processing.
 139  */
 140 int nf_bridge_copy_header(struct sk_buff *skb)
 141 {
 142         int err;
 143         int header_size = ETH_HLEN;
 144
 145         if (skb->protocol == htons(ETH_P_8021Q))
 146                 header_size += VLAN_HLEN;
 147
 148         err = skb_cow(skb, header_size);
 149         if (err)
 150                 return err;
 151
 152         skb_copy_to_linear_data_offset(skb, -header_size,
 153                                        skb->nf_bridge->data, header_size);
 154
 155         if (skb->protocol == htons(ETH_P_8021Q))
 156                 __skb_push(skb, VLAN_HLEN);
 157         return 0;
 158 }
 159
 160 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 161 /* Undo the changes made for ip6tables PREROUTING and continue the
 162  * bridge PRE_ROUTING hook. */
 163 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 164 {
 165         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 166
 167         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 168                 skb->pkt_type = PACKET_OTHERHOST;
 169                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 170         }
 171         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 172
 173         skb->dst = (struct dst_entry *)&__fake_rtable;
 174         dst_hold(skb->dst);
 175
 176         skb->dev = nf_bridge->physindev;
 177         if (skb->protocol == htons(ETH_P_8021Q)) {
 178                 skb_push(skb, VLAN_HLEN);
 179                 skb->network_header -= VLAN_HLEN;
 180         }
 181         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 182                        br_handle_frame_finish, 1);
 183
 184         return 0;
 185 }
 186
 187 static void __br_dnat_complain(void)
 188 {
 189         static unsigned long last_complaint;
 190
 191         if (jiffies - last_complaint >= 5 * HZ) {
 192                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 193                        "forwarding to be enabled\n");
 194                 last_complaint = jiffies;
 195         }
 196 }
 197
 198 /* This requires some explaining. If DNAT has taken place,
 199  * we will need to fix up the destination Ethernet address,
 200  * and this is a tricky process.
 201  *
 202  * There are two cases to consider:
 203  * 1. The packet was DNAT'ed to a device in the same bridge
 204  *    port group as it was received on. We can still bridge
 205  *    the packet.
 206  * 2. The packet was DNAT'ed to a different device, either
 207  *    a non-bridged device or another bridge port group.
 208  *    The packet will need to be routed.
 209  *
 210  * The correct way of distinguishing between these two cases is to
 211  * call ip_route_input() and to look at skb->dst->dev, which is
 212  * changed to the destination device if ip_route_input() succeeds.
 213  *
 214  * Let us first consider the case that ip_route_input() succeeds:
 215  *
 216  * If skb->dst->dev equals the logical bridge device the packet
 217  * came in on, we can consider this bridging. We then call
 218  * skb->dst->output() which will make the packet enter br_nf_local_out()
 219  * not much later. In that function it is assured that the iptables
 220  * FORWARD chain is traversed for the packet.
 221  *
 222  * Otherwise, the packet is considered to be routed and we just
 223  * change the destination MAC address so that the packet will
 224  * later be passed up to the IP stack to be routed. For a redirected
 225  * packet, ip_route_input() will give back the localhost as output device,
 226  * which differs from the bridge device.
 227  *
 228  * Let us now consider the case that ip_route_input() fails:
 229  *
 230  * This can be because the destination address is martian, in which case
 231  * the packet will be dropped.
 232  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 233  * will fail, while __ip_route_output_key() will return success. The source
 234  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 235  * thinks we're handling a locally generated packet and won't care
 236  * if IP forwarding is allowed. We send a warning message to the users's
 237  * log telling her to put IP forwarding on.
 238  *
 239  * ip_route_input() will also fail if there is no route available.
 240  * In that case we just drop the packet.
 241  *
 242  * --Lennert, 20020411
 243  * --Bart, 20020416 (updated)
 244  * --Bart, 20021007 (updated)
 245  * --Bart, 20062711 (updated) */
 246 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 247 {
 248         if (skb->pkt_type == PACKET_OTHERHOST) {
 249                 skb->pkt_type = PACKET_HOST;
 250                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 251         }
 252         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 253
 254         skb->dev = bridge_parent(skb->dev);
 255         if (!skb->dev)
 256                 kfree_skb(skb);
 257         else {
 258                 if (skb->protocol == htons(ETH_P_8021Q)) {
 259                         skb_pull(skb, VLAN_HLEN);
 260                         skb->network_header += VLAN_HLEN;
 261                 }
 262                 skb->dst->output(skb);
 263         }
 264         return 0;
 265 }
 266
 267 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 268 {
 269         struct net_device *dev = skb->dev;
 270         struct iphdr *iph = ip_hdr(skb);
 271         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 272         int err;
 273
 274         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 275                 skb->pkt_type = PACKET_OTHERHOST;
 276                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 277         }
 278         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 279         if (dnat_took_place(skb)) {
 280                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 281                         struct rtable *rt;
 282                         struct flowi fl = {
 283                                 .nl_u = {
 284                                         .ip4_u = {
 285                                                  .daddr = iph->daddr,
 286                                                  .saddr = 0,
 287                                                  .tos = RT_TOS(iph->tos) },
 288                                 },
 289                                 .proto = 0,
 290                         };
 291                         struct in_device *in_dev = in_dev_get(dev);
 292
 293                         /* If err equals -EHOSTUNREACH the error is due to a
 294                          * martian destination or due to the fact that
 295                          * forwarding is disabled. For most martian packets,
 296                          * ip_route_output_key() will fail. It won't fail for 2 types of
 297                          * martian destinations: loopback destinations and destination
 298                          * 0.0.0.0. In both cases the packet will be dropped because the
 299                          * destination is the loopback device and not the bridge. */
 300                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 301                                 goto free_skb;
 302
 303                         if (!ip_route_output_key(&rt, &fl)) {
 304                                 /* - Bridged-and-DNAT'ed traffic doesn't
 305                                  *   require ip_forwarding. */
 306                                 if (((struct dst_entry *)rt)->dev == dev) {
 307                                         skb->dst = (struct dst_entry *)rt;
 308                                         goto bridged_dnat;
 309                                 }
 310                                 /* we are sure that forwarding is disabled, so printing
 311                                  * this message is no problem. Note that the packet could
 312                                  * still have a martian destination address, in which case
 313                                  * the packet could be dropped even if forwarding were enabled */
 314                                 __br_dnat_complain();
 315                                 dst_release((struct dst_entry *)rt);
 316                         }
 317 free_skb:
 318                         kfree_skb(skb);
 319                         return 0;
 320                 } else {
 321                         if (skb->dst->dev == dev) {
 322 bridged_dnat:
 323                                 /* Tell br_nf_local_out this is a
 324                                  * bridged frame */
 325                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 326                                 skb->dev = nf_bridge->physindev;
 327                                 if (skb->protocol ==
 328                                     htons(ETH_P_8021Q)) {
 329                                         skb_push(skb, VLAN_HLEN);
 330                                         skb->network_header -= VLAN_HLEN;
 331                                 }
 332                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 333                                                skb, skb->dev, NULL,
 334                                                br_nf_pre_routing_finish_bridge,
 335                                                1);
 336                                 return 0;
 337                         }
 338                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 339                         skb->pkt_type = PACKET_HOST;
 340                 }
 341         } else {
 342                 skb->dst = (struct dst_entry *)&__fake_rtable;
 343                 dst_hold(skb->dst);
 344         }
 345
 346         skb->dev = nf_bridge->physindev;
 347         if (skb->protocol == htons(ETH_P_8021Q)) {
 348                 skb_push(skb, VLAN_HLEN);
 349                 skb->network_header -= VLAN_HLEN;
 350         }
 351         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 352                        br_handle_frame_finish, 1);
 353
 354         return 0;
 355 }
 356
 357 /* Some common code for IPv4/IPv6 */
 358 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 359 {
 360         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 361
 362         if (skb->pkt_type == PACKET_OTHERHOST) {
 363                 skb->pkt_type = PACKET_HOST;
 364                 nf_bridge->mask |= BRNF_PKT_TYPE;
 365         }
 366
 367         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 368         nf_bridge->physindev = skb->dev;
 369         skb->dev = bridge_parent(skb->dev);
 370
 371         return skb->dev;
 372 }
 373
 374 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 375 static int check_hbh_len(struct sk_buff *skb)
 376 {
 377         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 378         u32 pkt_len;
 379         const unsigned char *nh = skb_network_header(skb);
 380         int off = raw - nh;
 381         int len = (raw[1] + 1) << 3;
 382
 383         if ((raw + len) - skb->data > skb_headlen(skb))
 384                 goto bad;
 385
 386         off += 2;
 387         len -= 2;
 388
 389         while (len > 0) {
 390                 int optlen = nh[off + 1] + 2;
 391
 392                 switch (nh[off]) {
 393                 case IPV6_TLV_PAD0:
 394                         optlen = 1;
 395                         break;
 396
 397                 case IPV6_TLV_PADN:
 398                         break;
 399
 400                 case IPV6_TLV_JUMBO:
 401                         if (nh[off + 1] != 4 || (off & 3) != 2)
 402                                 goto bad;
 403                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 404                         if (pkt_len <= IPV6_MAXPLEN ||
 405                             ipv6_hdr(skb)->payload_len)
 406                                 goto bad;
 407                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 408                                 goto bad;
 409                         if (pskb_trim_rcsum(skb,
 410                                             pkt_len + sizeof(struct ipv6hdr)))
 411                                 goto bad;
 412                         nh = skb_network_header(skb);
 413                         break;
 414                 default:
 415                         if (optlen > len)
 416                                 goto bad;
 417                         break;
 418                 }
 419                 off += optlen;
 420                 len -= optlen;
 421         }
 422         if (len == 0)
 423                 return 0;
 424 bad:
 425         return -1;
 426
 427 }
 428
 429 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 430  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 431 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 432                                            struct sk_buff *skb,
 433                                            const struct net_device *in,
 434                                            const struct net_device *out,
 435                                            int (*okfn)(struct sk_buff *))
 436 {
 437         struct ipv6hdr *hdr;
 438         u32 pkt_len;
 439
 440         if (skb->len < sizeof(struct ipv6hdr))
 441                 goto inhdr_error;
 442
 443         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 444                 goto inhdr_error;
 445
 446         hdr = ipv6_hdr(skb);
 447
 448         if (hdr->version != 6)
 449                 goto inhdr_error;
 450
 451         pkt_len = ntohs(hdr->payload_len);
 452
 453         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 454                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 455                         goto inhdr_error;
 456                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 457                         goto inhdr_error;
 458         }
 459         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 460                 goto inhdr_error;
 461
 462         nf_bridge_put(skb->nf_bridge);
 463         if (!nf_bridge_alloc(skb))
 464                 return NF_DROP;
 465         if (!setup_pre_routing(skb))
 466                 return NF_DROP;
 467
 468         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 469                 br_nf_pre_routing_finish_ipv6);
 470
 471         return NF_STOLEN;
 472
 473 inhdr_error:
 474         return NF_DROP;
 475 }
 476
 477 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 478  * Replicate the checks that IPv4 does on packet reception.
 479  * Set skb->dev to the bridge device (i.e. parent of the
 480  * receiving device) to make netfilter happy, the REDIRECT
 481  * target in particular.  Save the original destination IP
 482  * address to be able to detect DNAT afterwards. */
 483 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 484                                       const struct net_device *in,
 485                                       const struct net_device *out,
 486                                       int (*okfn)(struct sk_buff *))
 487 {
 488         struct iphdr *iph;
 489         __u32 len;
 490         struct sk_buff *skb = *pskb;
 491
 492         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 493 #ifdef CONFIG_SYSCTL
 494                 if (!brnf_call_ip6tables)
 495                         return NF_ACCEPT;
 496 #endif
 497                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 498                         goto out;
 499
 500                 if (skb->protocol == htons(ETH_P_8021Q)) {
 501                         skb_pull_rcsum(skb, VLAN_HLEN);
 502                         skb->network_header += VLAN_HLEN;
 503                 }
 504                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 505         }
 506 #ifdef CONFIG_SYSCTL
 507         if (!brnf_call_iptables)
 508                 return NF_ACCEPT;
 509 #endif
 510
 511         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 512                 return NF_ACCEPT;
 513
 514         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 515                 goto out;
 516
 517         if (skb->protocol == htons(ETH_P_8021Q)) {
 518                 skb_pull_rcsum(skb, VLAN_HLEN);
 519                 skb->network_header += VLAN_HLEN;
 520         }
 521
 522         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 523                 goto inhdr_error;
 524
 525         iph = ip_hdr(skb);
 526         if (iph->ihl < 5 || iph->version != 4)
 527                 goto inhdr_error;
 528
 529         if (!pskb_may_pull(skb, 4 * iph->ihl))
 530                 goto inhdr_error;
 531
 532         iph = ip_hdr(skb);
 533         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 534                 goto inhdr_error;
 535
 536         len = ntohs(iph->tot_len);
 537         if (skb->len < len || len < 4 * iph->ihl)
 538                 goto inhdr_error;
 539
 540         pskb_trim_rcsum(skb, len);
 541
 542         nf_bridge_put(skb->nf_bridge);
 543         if (!nf_bridge_alloc(skb))
 544                 return NF_DROP;
 545         if (!setup_pre_routing(skb))
 546                 return NF_DROP;
 547         store_orig_dstaddr(skb);
 548
 549         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 550                 br_nf_pre_routing_finish);
 551
 552         return NF_STOLEN;
 553
 554 inhdr_error:
 555 //      IP_INC_STATS_BH(IpInHdrErrors);
 556 out:
 557         return NF_DROP;
 558 }
 559
 560
 561 /* PF_BRIDGE/LOCAL_IN ************************************************/
 562 /* The packet is locally destined, which requires a real
 563  * dst_entry, so detach the fake one.  On the way up, the
 564  * packet would pass through PRE_ROUTING again (which already
 565  * took place when the packet entered the bridge), but we
 566  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 567  * prevent this from happening. */
 568 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 569                                    const struct net_device *in,
 570                                    const struct net_device *out,
 571                                    int (*okfn)(struct sk_buff *))
 572 {
 573         struct sk_buff *skb = *pskb;
 574
 575         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 576                 dst_release(skb->dst);
 577                 skb->dst = NULL;
 578         }
 579
 580         return NF_ACCEPT;
 581 }
 582
 583 /* PF_BRIDGE/FORWARD *************************************************/
 584 static int br_nf_forward_finish(struct sk_buff *skb)
 585 {
 586         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 587         struct net_device *in;
 588
 589         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 590                 in = nf_bridge->physindev;
 591                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 592                         skb->pkt_type = PACKET_OTHERHOST;
 593                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 594                 }
 595         } else {
 596                 in = *((struct net_device **)(skb->cb));
 597         }
 598         if (skb->protocol == htons(ETH_P_8021Q)) {
 599                 skb_push(skb, VLAN_HLEN);
 600                 skb->network_header -= VLAN_HLEN;
 601         }
 602         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 603                        skb->dev, br_forward_finish, 1);
 604         return 0;
 605 }
 606
 607 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 608  * netfilter with indev and outdev set to the bridge device,
 609  * but we are still able to filter on the 'real' indev/outdev
 610  * because of the physdev module. For ARP, indev and outdev are the
 611  * bridge ports. */
 612 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 613                                      const struct net_device *in,
 614                                      const struct net_device *out,
 615                                      int (*okfn)(struct sk_buff *))
 616 {
 617         struct sk_buff *skb = *pskb;
 618         struct nf_bridge_info *nf_bridge;
 619         struct net_device *parent;
 620         int pf;
 621
 622         if (!skb->nf_bridge)
 623                 return NF_ACCEPT;
 624
 625         parent = bridge_parent(out);
 626         if (!parent)
 627                 return NF_DROP;
 628
 629         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 630                 pf = PF_INET;
 631         else
 632                 pf = PF_INET6;
 633
 634         if (skb->protocol == htons(ETH_P_8021Q)) {
 635                 skb_pull(*pskb, VLAN_HLEN);
 636                 (*pskb)->network_header += VLAN_HLEN;
 637         }
 638
 639         nf_bridge = skb->nf_bridge;
 640         if (skb->pkt_type == PACKET_OTHERHOST) {
 641                 skb->pkt_type = PACKET_HOST;
 642                 nf_bridge->mask |= BRNF_PKT_TYPE;
 643         }
 644
 645         /* The physdev module checks on this */
 646         nf_bridge->mask |= BRNF_BRIDGED;
 647         nf_bridge->physoutdev = skb->dev;
 648
 649         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 650                 br_nf_forward_finish);
 651
 652         return NF_STOLEN;
 653 }
 654
 655 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 656                                       const struct net_device *in,
 657                                       const struct net_device *out,
 658                                       int (*okfn)(struct sk_buff *))
 659 {
 660         struct sk_buff *skb = *pskb;
 661         struct net_device **d = (struct net_device **)(skb->cb);
 662
 663 #ifdef CONFIG_SYSCTL
 664         if (!brnf_call_arptables)
 665                 return NF_ACCEPT;
 666 #endif
 667
 668         if (skb->protocol != htons(ETH_P_ARP)) {
 669                 if (!IS_VLAN_ARP(skb))
 670                         return NF_ACCEPT;
 671                 skb_pull(*pskb, VLAN_HLEN);
 672                 (*pskb)->network_header += VLAN_HLEN;
 673         }
 674
 675         if (arp_hdr(skb)->ar_pln != 4) {
 676                 if (IS_VLAN_ARP(skb)) {
 677                         skb_push(*pskb, VLAN_HLEN);
 678                         (*pskb)->network_header -= VLAN_HLEN;
 679                 }
 680                 return NF_ACCEPT;
 681         }
 682         *d = (struct net_device *)in;
 683         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 684                 (struct net_device *)out, br_nf_forward_finish);
 685
 686         return NF_STOLEN;
 687 }
 688
 689 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 690  *
 691  * This function sees both locally originated IP packets and forwarded
 692  * IP packets (in both cases the destination device is a bridge
 693  * device). It also sees bridged-and-DNAT'ed packets.
 694  *
 695  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 696  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 697  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 698  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 699  * will be executed.
 700  */
 701 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 702                                     const struct net_device *in,
 703                                     const struct net_device *out,
 704                                     int (*okfn)(struct sk_buff *))
 705 {
 706         struct net_device *realindev;
 707         struct sk_buff *skb = *pskb;
 708         struct nf_bridge_info *nf_bridge;
 709
 710         if (!skb->nf_bridge)
 711                 return NF_ACCEPT;
 712
 713         nf_bridge = skb->nf_bridge;
 714         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 715                 return NF_ACCEPT;
 716
 717         /* Bridged, take PF_BRIDGE/FORWARD.
 718          * (see big note in front of br_nf_pre_routing_finish) */
 719         nf_bridge->physoutdev = skb->dev;
 720         realindev = nf_bridge->physindev;
 721
 722         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 723                 skb->pkt_type = PACKET_OTHERHOST;
 724                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 725         }
 726         if (skb->protocol == htons(ETH_P_8021Q)) {
 727                 skb_push(skb, VLAN_HLEN);
 728                 skb->network_header -= VLAN_HLEN;
 729         }
 730
 731         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 732                 br_forward_finish);
 733         return NF_STOLEN;
 734 }
 735
 736 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 737 {
 738         if (skb->protocol == htons(ETH_P_IP) &&
 739             skb->len > skb->dev->mtu &&
 740             !skb_is_gso(skb))
 741                 return ip_fragment(skb, br_dev_queue_push_xmit);
 742         else
 743                 return br_dev_queue_push_xmit(skb);
 744 }
 745
 746 /* PF_BRIDGE/POST_ROUTING ********************************************/
 747 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 748                                        const struct net_device *in,
 749                                        const struct net_device *out,
 750                                        int (*okfn)(struct sk_buff *))
 751 {
 752         struct sk_buff *skb = *pskb;
 753         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 754         struct net_device *realoutdev = bridge_parent(skb->dev);
 755         int pf;
 756
 757 #ifdef CONFIG_NETFILTER_DEBUG
 758         /* Be very paranoid. This probably won't happen anymore, but let's
 759          * keep the check just to be sure... */
 760         if (skb_mac_header(skb) < skb->head ||
 761             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 762                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 763                        "bad mac.raw pointer.\n");
 764                 goto print_error;
 765         }
 766 #endif
 767
 768         if (!nf_bridge)
 769                 return NF_ACCEPT;
 770
 771         if (!realoutdev)
 772                 return NF_DROP;
 773
 774         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 775                 pf = PF_INET;
 776         else
 777                 pf = PF_INET6;
 778
 779 #ifdef CONFIG_NETFILTER_DEBUG
 780         if (skb->dst == NULL) {
 781                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 782                 goto print_error;
 783         }
 784 #endif
 785
 786         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 787          * about the value of skb->pkt_type. */
 788         if (skb->pkt_type == PACKET_OTHERHOST) {
 789                 skb->pkt_type = PACKET_HOST;
 790                 nf_bridge->mask |= BRNF_PKT_TYPE;
 791         }
 792
 793         if (skb->protocol == htons(ETH_P_8021Q)) {
 794                 skb_pull(skb, VLAN_HLEN);
 795                 skb->network_header += VLAN_HLEN;
 796         }
 797
 798         nf_bridge_save_header(skb);
 799
 800 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 801         if (nf_bridge->netoutdev)
 802                 realoutdev = nf_bridge->netoutdev;
 803 #endif
 804         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 805                 br_nf_dev_queue_xmit);
 806
 807         return NF_STOLEN;
 808
 809 #ifdef CONFIG_NETFILTER_DEBUG
 810 print_error:
 811         if (skb->dev != NULL) {
 812                 printk("[%s]", skb->dev->name);
 813                 if (realoutdev)
 814                         printk("[%s]", realoutdev->name);
 815         }
 816         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 817                skb->data);
 818         dump_stack();
 819         return NF_ACCEPT;
 820 #endif
 821 }
 822
 823 /* IP/SABOTAGE *****************************************************/
 824 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 825  * for the second time. */
 826 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 827                                    const struct net_device *in,
 828                                    const struct net_device *out,
 829                                    int (*okfn)(struct sk_buff *))
 830 {
 831         if ((*pskb)->nf_bridge &&
 832             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 833                 return NF_STOP;
 834         }
 835
 836         return NF_ACCEPT;
 837 }
 838
 839 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 840  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 841  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 842  * ip_refrag() can return NF_STOLEN. */
 843 static struct nf_hook_ops br_nf_ops[] = {
 844         { .hook = br_nf_pre_routing,
 845           .owner = THIS_MODULE,
 846           .pf = PF_BRIDGE,
 847           .hooknum = NF_BR_PRE_ROUTING,
 848           .priority = NF_BR_PRI_BRNF, },
 849         { .hook = br_nf_local_in,
 850           .owner = THIS_MODULE,
 851           .pf = PF_BRIDGE,
 852           .hooknum = NF_BR_LOCAL_IN,
 853           .priority = NF_BR_PRI_BRNF, },
 854         { .hook = br_nf_forward_ip,
 855           .owner = THIS_MODULE,
 856           .pf = PF_BRIDGE,
 857           .hooknum = NF_BR_FORWARD,
 858           .priority = NF_BR_PRI_BRNF - 1, },
 859         { .hook = br_nf_forward_arp,
 860           .owner = THIS_MODULE,
 861           .pf = PF_BRIDGE,
 862           .hooknum = NF_BR_FORWARD,
 863           .priority = NF_BR_PRI_BRNF, },
 864         { .hook = br_nf_local_out,
 865           .owner = THIS_MODULE,
 866           .pf = PF_BRIDGE,
 867           .hooknum = NF_BR_LOCAL_OUT,
 868           .priority = NF_BR_PRI_FIRST, },
 869         { .hook = br_nf_post_routing,
 870           .owner = THIS_MODULE,
 871           .pf = PF_BRIDGE,
 872           .hooknum = NF_BR_POST_ROUTING,
 873           .priority = NF_BR_PRI_LAST, },
 874         { .hook = ip_sabotage_in,
 875           .owner = THIS_MODULE,
 876           .pf = PF_INET,
 877           .hooknum = NF_IP_PRE_ROUTING,
 878           .priority = NF_IP_PRI_FIRST, },
 879         { .hook = ip_sabotage_in,
 880           .owner = THIS_MODULE,
 881           .pf = PF_INET6,
 882           .hooknum = NF_IP6_PRE_ROUTING,
 883           .priority = NF_IP6_PRI_FIRST, },
 884 };
 885
 886 #ifdef CONFIG_SYSCTL
 887 static
 888 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 889                             void __user * buffer, size_t * lenp, loff_t * ppos)
 890 {
 891         int ret;
 892
 893         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 894
 895         if (write && *(int *)(ctl->data))
 896                 *(int *)(ctl->data) = 1;
 897         return ret;
 898 }
 899
 900 static ctl_table brnf_table[] = {
 901         {
 902                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 903                 .procname       = "bridge-nf-call-arptables",
 904                 .data           = &brnf_call_arptables,
 905                 .maxlen         = sizeof(int),
 906                 .mode           = 0644,
 907                 .proc_handler   = &brnf_sysctl_call_tables,
 908         },
 909         {
 910                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 911                 .procname       = "bridge-nf-call-iptables",
 912                 .data           = &brnf_call_iptables,
 913                 .maxlen         = sizeof(int),
 914                 .mode           = 0644,
 915                 .proc_handler   = &brnf_sysctl_call_tables,
 916         },
 917         {
 918                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 919                 .procname       = "bridge-nf-call-ip6tables",
 920                 .data           = &brnf_call_ip6tables,
 921                 .maxlen         = sizeof(int),
 922                 .mode           = 0644,
 923                 .proc_handler   = &brnf_sysctl_call_tables,
 924         },
 925         {
 926                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 927                 .procname       = "bridge-nf-filter-vlan-tagged",
 928                 .data           = &brnf_filter_vlan_tagged,
 929                 .maxlen         = sizeof(int),
 930                 .mode           = 0644,
 931                 .proc_handler   = &brnf_sysctl_call_tables,
 932         },
 933         { .ctl_name = 0 }
 934 };
 935
 936 static ctl_table brnf_bridge_table[] = {
 937         {
 938                 .ctl_name       = NET_BRIDGE,
 939                 .procname       = "bridge",
 940                 .mode           = 0555,
 941                 .child          = brnf_table,
 942         },
 943         { .ctl_name = 0 }
 944 };
 945
 946 static ctl_table brnf_net_table[] = {
 947         {
 948                 .ctl_name       = CTL_NET,
 949                 .procname       = "net",
 950                 .mode           = 0555,
 951                 .child          = brnf_bridge_table,
 952         },
 953         { .ctl_name = 0 }
 954 };
 955 #endif
 956
 957 int __init br_netfilter_init(void)
 958 {
 959         int ret;
 960
 961         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 962         if (ret < 0)
 963                 return ret;
 964 #ifdef CONFIG_SYSCTL
 965         brnf_sysctl_header = register_sysctl_table(brnf_net_table);
 966         if (brnf_sysctl_header == NULL) {
 967                 printk(KERN_WARNING
 968                        "br_netfilter: can't register to sysctl.\n");
 969                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 970                 return -ENOMEM;
 971         }
 972 #endif
 973         printk(KERN_NOTICE "Bridge firewalling registered\n");
 974         return 0;
 975 }
 976
 977 void br_netfilter_fini(void)
 978 {
 979         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 980 #ifdef CONFIG_SYSCTL
 981         unregister_sysctl_table(brnf_sysctl_header);
 982 #endif
 983 }